1. Levinson’s Theorem for
Scattering on Graphs
DJ Strouse
University of Southern California
Andrew M. Childs
University of Waterloo
2. Why Scatter on Graphs?
• NAND Tree problem:
• Best classical algorithm:
– Randomized
– Only needs to evaluate of the leaves
Figure from: Farhi E, Goldstone J, Gutmann S. A Quantum Algorithm for the
Hamiltonian NAND Tree.
3. Why Scatter on Graphs?
Can a quantum algorithm do better?
4. Why Scatter on Graphs?
• Farhi, Goldstone, Gutmann (2007)
• Connections to parallel nodes represent input
• Prepare a traveling wave packet on the left…
• …let it loose…
• …if found on the right after fixed time, answer 1
Figure from: Farhi E, Goldstone J, Gutmann S. A Quantum Algorithm for the
Hamiltonian NAND Tree.
5. Why Scatter on Graphs?
Can scattering on graphs offer a quantum speedup on
other interesting problems?
6. What You Skipped the BBQ For
• Goal: Relate (a certain property of) scattering states
(called the winding of their phase) to the number of
bound states and the size of a graph
1. Crash Course in Graphs
2. Introduction to Quantum Walks (& scattering on graphs)
3. Meet the Eigenstates
i. Scattering states (& the winding of their phase)
ii. (The many species of) bound states
4. Some Examples: explore relation between winding & bound states
5. A Brief History of Levinson’s Theorem
6. Explain the Title: “Levinson’s Theorem for Scattering on Graphs”
7. A Sketch of the Proof
8. A Briefer Future of Levinson’s Theorem
10. Meet the Eigenstates
Resolve the
identity…
Scattering states Bound states
Diagonalize the
Hamiltonian…
Represent your
favorite state…
…and evolve it!
15. Standard & Alternating Bound States
• SBS: exponentially • ABS: same as SBS but
decaying amplitude on with alternating sign
the tail
Exist at discrete κ depending on graph structure
16. Confined Bound States
• Eigenstates that live entirely on the graph
Eigenstate of G
with zero
amplitude on the
attachment point
Exist at discrete E depending on graph structure
17. Standard & Alternating
Half-Bound States
• HBS: constant • AHBS: same as HBS with
amplitude on the tail alternating sign
•Unnormalizable like SS… but obtainable from BS eqns
•Energy wedged between SBS/ABS and SS
•May or may not exist depending on graph structure
19. Into the Jungle:
Bound States & Phase Shifts in the Wild
One SBS & One ABS One HBS & One AHBS
One SBS, One ABS, & One CBS No BS
20. A Brief History of Levinson’s Theorem
• Continuum Case:
– Levinson (1949)
– No CBS
Potential on a half-line
(modeling spherically
symmetric 3D potential)
Excerpt from: Dong S-H and Ma Z-Q 2000 Levinson's theorem for the Schrödinger
equation in one dimension Int. J. Theor. Phys. 39 469-81
21. A Brief History of Levinson’s Theorem
• Continuum Case:
– Levinson (1949)
– No CBS
• Discrete Case:
– Case & Kac (1972)
• Graph = chain with self-loops
• No CBS & ignored HBS
– Hinton, Klaus, & Shaw (1991)
• Included HBS
• …but still just chain with self-loops
30. Into the Jungle:
Bound States & Phase Shifts in the Wild
One SBS & One ABS One HBS & One AHBS
One SBS, One ABS, & One CBS No BS
31. Future Work
• What about multiple tails?
– Now R is a matrix (called the S-matrix)…
– The generalized argument principle is not so elegant…
Excerpt from: H. Ammari, H. Kang, and H. Lee, Layer Potential Techniques in Spectral Analysis, Mathematical
Surveys and Monographs, Vol. 153, American Mathematical Society, Providence RI, 2009.
32. Future Work
• What about multiple tails?
– Now R is a matrix (called the S-matrix)…
– The generalized argument principle is not so elegant…
• Possible step towards new quantum algorithms?
– Are there interesting problems that can be couched in terms of
the number of bound states and vertices of a graph?
– What properties of graphs make them nice habitats for the
various species of bound states?
Editor's Notes
Decision problem
Worth understanding the scattering scene a little better
Worth understanding the scattering scene a little better
This talk spans physics and CS, so each of you is absolved from being expected to know anything. (there are no dumb questions)
In general in scattering picture, directed graphs with weights to & from complex conjugates of one another, so adjacency matrix is Hermitian, but we simplify to the case with real, symmetric weights. Everything we’ll discuss carries over to the more general case.
If you know a bit of QM, then you know that to talk about things moving around, we need a Hilbert space and a Hamiltonian.We’ll only need notation for “tail” basis states.
Long graphs aren’t very interesting. We need a tail to speak of scattering.Most general case = multiple tails (universal for quantum computation)We focus on single tail, so “scattering” involves only a phase shift.With tail, adjacency matrix is now infinite!
SS normalizable & one speciesBS unnormalizable & many speciesSo this is why we even care about theorems relating scattering and bound states
Calculations on tails fixes energyCalculations on graphs fix amplitudes on graph (just more linear equations)
Phi(k) is the property of SS that we’re interested in. More specifically, we’re interested in the winding of Phi(k).
Zero amplitude on entire tail, including attachment point
Some of you will spot an unfortunate pun on this page.
Stress main differences between SS & BS:NormalizabilityLikely locationExistence
Suggests connectionBased on these examples, you may want to take a guess at what the theorem will look like
Levinson paper is in another language – both mathematically and linguisticallyLevinson in continuum with no CBS and momenta unboundedOther discrete: chains with self-loops & no CBS
Levinson paper is in another language – both mathematically and linguisticallyLevinson in continuum with no CBS and momenta unboundedOther discrete: chains with self-loops & no CBS
And just in case you don’t have a photographic memory, here’s the definitions.Did anyone actually guess this form from the examples?Very odd: bound states & scattering states – orthogonal (expect nothing to do with each other) but somehow related
AC: Can we just venture off the unit circle into the complex plane like this? Yes, because analytic functions are uniquely defined for this extension.AP: behavior of function inside border tells you something about behavior on border (applies to functions that are analytic except at a finite number of points)
AC: Can we just venture off the unit circle into the complex plane like this? Yes, because analytic functions are uniquely defined for this extension.AP: behavior of function inside border tells you something about behavior on border (applies to functions that are analytic except at a finite number of points)
AC: Can we just venture off the unit circle into the complex plane like this? Yes, because analytic functions are uniquely defined for this extension.AP: behavior of function inside border tells you something about behavior on border (applies to functions that are analytic except at a finite number of points)
AC: Can we just venture off the unit circle into the complex plane like this? Yes, because analytic functions are uniquely defined for this extension.AP: behavior of function inside border tells you something about behavior on border (applies to functions that are analytic except at a finite number of points)
AC: Can we just venture off the unit circle into the complex plane like this? Yes, because analytic functions are uniquely defined for this extension.AP: behavior of function inside border tells you something about behavior on border (applies to functions that are analytic except at a finite number of points)
AC: Can we just venture off the unit circle into the complex plane like this? Yes, because analytic functions are uniquely defined for this extension.AP: behavior of function inside border tells you something about behavior on border (applies to functions that are analytic except at a finite number of points)
AC: Can we just venture off the unit circle into the complex plane like this? Yes, because analytic functions are uniquely defined for this extension.AP: behavior of function inside border tells you something about behavior on border (applies to functions that are analytic except at a finite number of points)
Suggests connectionBased on these examples, you may want to take a guess at what the theorem will look like